Microsoft Licensed Weights
2 pasos1 piezas2 Fuentes
Microsoft has introduced "MAI Licensed Weights," a program allowing enterprises to customize its frontier AI models with proprietary data while retaining control over deployment and intellectual property within Azure. Announced at the Ignite conference on November 17, 2026, this initiative positions itself as a middle ground between fully closed proprietary models (like those from OpenAI and Anthropic) and fully open-weight models (like Mistral's). Under this scheme, customers can adapt Microsoft AI (MAI) models to their specific workflows without exposing their core IP to third-party training pools unless explicitly agreed upon. Mustafa Suleyman, CEO of MAI, emphasized that businesses would no longer need to "pour their IQ into someone else’s model." The program is delivered through Frontier Company engagements, requiring direct contact with Microsoft account representatives rather than public download links. Early case studies cited by Microsoft include AT&T using customized MAI Code 2 Flash for 5G network analysis at 1/50th the cost of GPT-6.1 Sol, and Alaska Airlines improving maintenance report classification accuracy from 81.2% to 91.1%. This move challenges the prevailing binary choice in enterprise AI: high-cost, non-customizable closed models versus lower-capability, customizable open weights.
Cómo llegó hasta aquí
November 17, 2026: Microsoft announces MAI Licensed Weights during the Ignite keynote, detailing how enterprises can secure and customize frontier models in Azure.
microsoft.ai ↗November 18, 2026: ZDNET reports on the launch, highlighting the strategic positioning against OpenAI and Anthropic and citing customer performance metrics.
zdnet.com ↗